Object Position Estimation Using Probability Density Functions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object position estimation systems face challenges in expanding the estimation range while maintaining accuracy, particularly when stationary objects occlude the target object, leading to erroneous detection and decreased accuracy.
Innovation Solution
The system employs image acquisition means from multiple directions, data processing to create object region images, score calculation, view status determination, and probability density functions to estimate object position, even in non-overlapping fields, by accounting for stationary objects and their occlusion effects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If the estimation range is expanded to include non-overlapping regions of multiple cameras, then the coverage area increases, but erroneous detection occurs and measurement precision decreases
Solution Approach 1:
The system performs preliminary calibration to obtain camera parameters and pre-establishes the relationship between real space and image space for each camera. This preliminary action enables the system to accurately determine whether detected objects fall within valid estimation regions, allowing expansion of the estimation range while maintaining detection accuracy through pre-computed spatial relationships.
Solution Approach 2:
The system introduces camera parameters and coordinate transformation relationships as intermediary elements between the raw image data and object position estimation. By using these intermediaries to map detected objects back to real space coordinates and verify their validity, the system can safely expand the estimation range to non-overlapping regions while preventing erroneous detections through mathematical verification.
2Reliability
If stationary objects are present that conceal the target object, then the object position estimation becomes difficult, but expanding camera coverage can help detect the target object
Solution Approach 1:
The system transitions from two-dimensional image plane analysis to three-dimensional real space coordinate analysis by applying camera parameters and coordinate transformations. This dimensional change allows the system to determine whether stationary objects are actually concealing the target object by checking spatial relationships in 3D space, thereby improving detection reliability without requiring additional cameras or complex multi-view geometry.
Solution Approach 2:
The system changes the parameter space from image coordinates to real space coordinates using camera calibration parameters. By transforming the detection problem into the real space parameter domain, the system can accurately assess whether stationary objects are causing occlusion and make reliable position estimations even when the target object is partially concealed in the image plane.
3Measurement precision
If only overlapping regions of all cameras are used for position estimation, then measurement precision is maintained, but the estimation range is limited
Solution Approach 1:
The system performs preliminary calibration to establish accurate camera parameters and coordinate transformation relationships before the actual object detection process. This pre-computed spatial information enables the system to confidently expand the estimation range to non-overlapping regions while maintaining position estimation accuracy through mathematically verified coordinate mappings.
Solution Approach 2:
The system uses coordinate transformation to map detection results from the two-dimensional image plane to three-dimensional real space. This dimensional change allows the system to determine the validity of position estimates in non-overlapping regions by checking whether the transformed coordinates fall within the valid field of view of at least one camera, thereby expanding coverage while maintaining accuracy.
Data Source
Figure 1
Figure 2(a)~2(b)
Figure 3
AI summary
Score determination means 62 determines a score, which indicates how accurately points in real space in respective images respectively acquired by a plurality of image acquisition means represent a target object, for each combination of image acquisition means and point in real space. View status determination means 63 determines the status of view to each point in real space from each image acquisition means by using non-target object presence information indicating to image acquisition means where a non-target object that conceals a target object is present, and determines probability density functions of a score corresponding to the result of the view status determination. Existence probability calculation means 64 determines an existence probability of a target object being present at a point in real space by using probability density functions for each combination of the image acquisition means and point in real space. Estimation information creation means 65 creates information for estimating position from the existence probability.